Yeah and that's the thing - the design talent exists in these communities, it's just historically been undervalued compared to the engineering. When projects actually bring designers in early instead of as an afterthought, the gap closes fast.
The missing piece is that "easy" and "respects you" aren't actually opposites. They just got bundled together because the companies with UX budgets were also the ones monetising your data. That's a funding problem, not a design law.
We built Meos because your most honest thoughts deserve to stay yours. getmeos.com
Hmm, I'd push back slightly on hardware being the main barrier though. Even now you can run decent 7-8b models on a phone with 8GB RAM. The real bottleneck is tooling and context management around those smaller models, not raw parameter count.
Curious what kind of inference speeds you're seeing with 26B on a Mini. The 12B guide had the keep-alive and preload sorted nicely, but at 26B I'd expect memory pressure to get interesting with unified memory. Are you hitting swap at all?
The bit that gets me is how the AT Protocol basically made "owning your data" a spectrum instead of a binary choice. You can go full Blacksky or just run a PDS and call it a day. What's your take on where the sweet spot is between convenience and control?
Running inference on a dedicated peripheral device is such a good call for keeping the audio pipeline isolated. What are you using for the LLM runtime on it, and how's the power draw when it's generating playlists?
So we built a knowledge graph that lives on your device. Semantic neighbours, evolution history, temporal patterns. Ask a question and parallel curators trace connections across everything you've ever written. With cryptographic provenance on every answer. getmeos.com
We built search that works by meaning, not filenames. Your entries form a knowledge graph on-device. Ask a question, get answers from your own writing with citations. getmeos.com
The "good enough" tool you keep coming back to tells you something about what you actually need. We spent a long time asking that question before building Meos. Turns out the answer was: local storage, sync that works, and AI that knows your context. getmeos.com
We built that missing piece. Semantic search over your own writing, on your phone. Find old ideas by describing what they were about, not guessing which words you used. getmeos.com
We built retrieval that fuses semantic meaning, keyword match, and temporal signals on-device. You search by describing what you meant, not remembering exact words. getmeos.com
Done 👍
The bit that gets me is how invisible the loss is. You don't feel yourself getting weaker at something. You just... stop reaching for the muscle. And then one day you need it and it's not there. How are you managing that balance yourself?
The gap between "AI shoved into a product's UI" and "AI as a tool you actually direct" is massive. Consumer AI feels like someone else's idea of helpful. What's your workflow with CC look like day to day?
The hardest part I've found with any second brain setup is retrieval. Capturing is easy, organising is satisfying, but actually pulling up the right idea at the right moment months later? That's where most systems quietly fall apart.
Hmm, I'd actually push back a bit on MyMind for that though. It's great for saving stuff, but the retrieval is still keyword-ish. The real missing piece is searching by meaning across your own evolving ideas, not just bookmarks.
BlueSky as a dumping ground plus scattered docs is honestly how most people's "system" starts. The gap is always retrieval, finding that thing you know you wrote six months ago.
Every tool you rely on has a terms of service update waiting to change the rules.
Your notes in Meos don't have terms. They have no server. No account. No leverage over you.
Just your thoughts, on your device, fully yours.
When did that become radical?
Twelve tabs open. Four apps. Two docs you forgot existed. And somewhere in all that noise, the idea that would've changed the project.
What if the cost of tool sprawl isn't time... it's the thoughts you never reconnected?
That feature list creeping up on you is the real killer. You start with "just notes" and suddenly you're building half of Todoist and a markdown parser at 2am. Obsidian eating those problems for you is the right call.
The thing you're describing though, a tool that tracks how your thinking evolves and lets you sense-check against your own past ideas, that's a genuinely solvable problem. The issue is most tools try to be the thinker instead of just being a reliable memory. What methods are you using now?
Hmm, I'd push back a bit on "own your profiles" as the fix. The profile itself is the problem. Any platform where you are the product will find new ways to extract. What does your actual offline system for keeping track of contacts and networking notes look like?
I went through the same loop with a custom task tracker. Built it for months, learned a ton about local storage and sync, then quietly went back to a plain markdown folder. The "well enough" realisation is honestly underrated. What was the bit of Obsidian that finally won you back?
The cruel bit is that the hiatus used to mean something because people had longer memory. Now the algorithm buries you the second you stop feeding it. Do you find it changes how you approach making work, like feeling pressured to always be "on"?
When was the last time you searched your notes and found something better than what you were looking for?
Every tool you add is a tiny bet that this one will finally be the last one.
It never is.
What if the problem isn't finding the right tool. It's that your thoughts are scattered across twelve of them, and none of them talk to each other.
Experiment: I searched my notes for "that paper about memory and place" and it found a highlight I saved eleven months ago from an article I'd completely forgotten reading.
The pile isn't the problem. The silence is.
What if your notes could answer back?
Export/import is a solid starting point. When I built device sync I underestimated how gnarly conflict resolution gets once two devices edit the same item offline. CRDTs saved me there.
You didn't stop reading. You stopped being able to find what you read.
Every article you saved knew something. Every highlight had a reason. The problem was never your curiosity. It was that the pile had no memory.
What if it did?